A Dataset on Generative AI-Assisted Design, Academic Emotions, Metacognition, and Creative Performance in College Students
Description
Research Purpose and Design This dataset was generated from a quasi-experimental study exploring relationships among Generative Artificial Intelligence (GAI) tool use, academic emotions, metacognition, and creative task performance in design education. The study adopted an exploratory approach to examine whether the availability of an AI assistant was associated with variations in emotional experiences, performance outcomes, and the predictive roles of metacognitive components. Rather than testing directional hypotheses, it investigated potential patterns and interactions that could inform future theory development in human-AI collaborative learning. Data were collected in spring 2026 from 76 undergraduate students in an Advertising Poster Design course, divided into a GAI-assisted group (n=39) and a no-tool control group (n=37) through stratified quasi-random assignment. Data Collection and Variables All participants completed a 100-minute campus culture poster design task. The GAI group had access to Doubao, while the control group used no electronic devices. Before the task, students completed a metacognitive awareness inventory measuring metacognitive knowledge and regulation. Immediately after, they retrospectively rated eight academic emotions (enjoyment, hope, pride, anger, anxiety, shame, hopelessness, boredom) on a 0-100 scale. Task outputs were independently scored by two blind raters across effectiveness, correctness, and novelty. The dataset includes complete records for all participants, covering group assignment, metacognitive subscale scores, emotion ratings, and performance scores. Observed Patterns and Interpretive Notes Exploratory analyses reveal several notable patterns. GAI use appears associated with higher performance scores and elevated positive emotions, while negative emotions show less consistent group differences. Negative correlations between certain unpleasant emotions and novelty seem to emerge only in the non-GAI group, and the predictive strength of metacognitive knowledge on performance appears to differ between groups, suggesting that GAI tools may interact with affective and cognitive processes in ways that warrant further investigation. These observations are exploratory and do not imply causation; they point to potential associations for future confirmatory studies. Users should interpret the data with the following considerations: emotion measures are retrospective and reflect global experiences rather than dynamic fluctuations; grouping is based on quasi-randomisation rather than full random assignment; performance criteria are domain-specific to advertising poster design and may require adaptation; and metacognition and emotions were measured at different time points. The dataset is suited for secondary analyses such as correlation, moderation, or latent profile analyses, and may be combined with qualitative design rationales to deepen understanding of human-AI collaboration in creative tasks.
Files
Steps to reproduce
Data Generation and Research Procedures This dataset was produced through a quasi‑experimental study conducted in the spring semester of 2026 at a university visual communication design laboratory. The research employed a posttest‑only control‑group design. A total of 76 undergraduate students enrolled in an Advertising Poster Design course participated and were assigned to either a GAI‑assisted group (n=39) or a no‑tool control group (n=37) through stratified quasi‑random assignment based on baseline academic metrics. The experimental procedure was standardised as follows: all students first completed a metacognitive awareness inventory. They then engaged in a 100‑minute creative problem‑solving task—designing a poster for a university campus culture festival. The GAI group had access to Doubao, a multimodal AI tool, while the control group completed the task without any electronic devices. All sessions were conducted in physically separated spaces with restricted network access to ensure experimental control. The same instructor delivered standardised task instructions to both groups to minimise experimenter expectancy effects. Immediately after task completion, students retrospectively rated their academic emotions across eight dimensions using a 0–100 self‑report scale. Task outputs (poster designs and accompanying design rationales) were independently scored by two blind raters using a standardised rubric developed specifically for this study. The rubric evaluates performance along three dimensions—effectiveness, correctness, and novelty—with detailed behavioural anchors and scoring criteria for each level. The dataset comprises complete records for all 76 participants, including group assignment, metacognitive scores, emotion ratings, and performance scores. No specialised reagents or software were required beyond the AI tool and standard data processing tools.